Evaluation of metabolite biomarker candidates in detecting HCC in patients with liver cirrhosis

Abstract

Introduction

Hepatocellular carcinoma (HCC), the most prevalent form of liver cancer, ranks as the third leading cause of mortality globally. Patients diagnosed with HCC exhibit a dismal prognosis, mostly due to the emergence of symptoms in the advanced stages of the disease. Moreover, conventional biomarkers demonstrate insufficient efficacy in the early detection of HCC, hence highlighting the need for the identification of novel and more effective biomarkers. This study aims to evaluate a selected panel of serum biomarker candidates for the detection of HCC in patients with liver cirrhosis (CIRR). This is accomplished by targeted quantitation of the candidates using a triple quadrupole mass spectrometer.

Methods

Serum samples from 50 HCC cases (27 Stage I HCC), 50 patients with CIRR, and 25 healthy controls were analyzed using ultra-high-performance liquid chromatography-TSQ Altis Plus triple quadrupole mass spectrometry (UHPLC-MS/MS) by multiple reaction monitoring (MRM). Absolute quantification of 13 endogenous metabolites selected from previous studies was performed using the surrogate matrix approach by creating calibration curves for each metabolite. Statistical analyses included univariate testing with false discovery rate (FDR) correction, multivariable logistic regression adjusted for clinical covariates, and receiver operating characteristic (ROC) curves.

Results

Six metabolites primarily involving amino acid and bile acid metabolism were significantly altered in HCC vs. CIRR, with four of these also significant in Stage I HCC vs. CIRR. While AFP alone achieved AUCs of 0.773 ± 0.106 in HCC vs. CIRR and 0.804 ± 0.093 in Stage I HCC vs. CIRR. The combination of AFP with a six-metabolite panel improved discrimination (AUCs 0.870 ± 0.083 and 0.877 ± 0.059, respectively). Among the six metabolites, ornithine and proline remained associated with HCC after adjusting for confounding factors such as age, sex, BMI, MELD score, and HCV status.

Conclusion

Targeted metabolomics reveals reproducible metabolic alterations in HCC, including early-stage disease; however, substantial overlap with cirrhosis limits their independent diagnostic utility. Integration with AFP provides modest improvement, supporting a complementary multi-marker approach for HCC detection.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11306-026-02470-0.

Introduction

Hepatocellular carcinoma (HCC) is the most prevalent liver cancer and ranks among the top three causes of cancer-related mortality worldwide, with an annual incidence projected to exceed one million by 2025 (Zhan et al., 2025). Early-stage diagnosis remains a persistent challenge, as most patients present with advanced disease due to silent progression, especially among those with longstanding liver cirrhosis (CIRR), which is a known precursor in over 90% of HCC cases (Schulze et al., 2025).

Traditional serological and imaging approaches, primarily relying on alpha-fetoprotein (AFP), are limited in sensitivity and specificity for early HCC detection. AFP detects only 20–50% of early HCC cases and may be elevated in chronic liver disease independent of malignancy (Kim et al., 2023, Marrero et al., 2018, Wong et al., 2015). Other biomarkers, including des-gamma-carboxy prothrombin, glypican-3, novel metabolic signatures, and circulating metabolites, offer incremental improvements but do not yet provide adequate discrimination in the earliest disease stages based on reports from systematic reviews and meta-analyses (Wu et al., 2020, Li et al., 2018, Lu et al., 2022, Li et al., 2014).

Recent advances in high-throughput metabolomics and multi-omics integration have transformed biomarker discovery for HCC (Li et al., 2018, Bernhardt et al., 2025). Metabolomics offers detailed insights into metabolic reprogramming a hallmark of cancer, by profiling small molecules whose abundance reflects underlying biological changes in hepatocarcinogenesis (Feng et al., 2022). Emerging metabolite panels, especially those incorporating amino acids (e.g., proline, ornithine, valine), tricarboxylic acid (TCA) cycle intermediates (e.g., succinic acid, fumaric acid, malic acid), and lipids, have demonstrated superior diagnostic accuracy over conventional markers in global cohort studies (Al-Amodi and Kamel 2024, Morine et al., 2022).

High specificity and sensitivity achieved with targeted quantitation platforms, such as triple quadrupole mass spectrometry, underscore the promise of metabolomic liquid biopsies for early-stage HCC (Rashid et al., 2023). However, the clinical utility of these biomarkers relies on robust validation through large, diverse cohorts and integration with genomic and proteomic signatures to capture the full complexity of HCC pathogenesis.

Thus, this study aims to rigorously quantify a selected panel of serum biomarker candidates identified in prior untargeted metabolomics analyses and literature for the detection of HCC in patients with CIRR. Targeted quantification is conducted using multiple reaction monitoring (MRM) on a triple quadrupole mass spectrometer, with all selected metabolites verified against commercially available standards by retention time, mass-to-charge (m/z), and MS/MS fragment comparison. The study analyzes serum samples from 50 HCC cases, 50 cirrhosis patients, and 25 healthy controls with ultra-high-performance liquid chromatography-TSQ Altis Plus triple quadrupole mass spectrometry (UHPLC-MS/MS). Absolute quantification of each endogenous metabolite is achieved via the surrogate matrix calibration curve method. Significant alterations are observed in the concentration of amino acids, including ornithine, valine, isoleucine, proline, glycoursodeoxycholic acid (GUDCA), and phenylalanine in HCC vs. CIRR, supporting the notion that targeted metabolomics enables sensitive identification of candidate biomarkers for HCC detection in high-risk patients with liver cirrhosis.

Materials and methods

Overview

Figure 1 illustrates the overall workflow of our MRM-based targeted quantitation of metabolite biomarker candidates in serum samples from 50 HCC cases, 50 patients with CIRR, and 25 healthy volunteers. This targeted study is a continuation of our previously published untargeted study, where we reported a panel of putative metabolite biomarkers for the identification of HCC in patients with CIRR. In this study, we quantified 13 metabolites selected from our previous untargeted analysis and reported in the literature. Metabolites were selected based on their statistical relevance, biological importance in HCC (particularly amino acid and bile acid metabolism), and the availability of commercially available reference standards required for accurate quantification. Additional metabolites, including fatty acids and lipid species identified in our discovery phase, were not included due to the need for different extraction and analytical workflows and will be investigated in the future separately using lipid-focused analytical approaches. The same LC parameters previously used in our untargeted study were used for this targeted study. However, MS conditions, including the MRM transitions, collision energy, spray voltage, and ionization mode, were individually optimized in order to obtain better sensitivity and detection. Proper chromatographic resolution is required to separate isomeric metabolites such as leucine and isoleucine without compromising peak shape and accurate peak integration due to the insufficient mass spectrometric selectivity. To address this, separate MRM transitions were chosen for these two isomeric amino acids, which improves their peak resolution and integration. Likewise, separate MRM transitions were chosen for two bile acids (BAs), including glycodeoxycholic acid (GDCA) and GUDCA, despite having exact m/z and fragmentationpatterns.

Fig. 1.

Workflow of our MRM-based targeted quantitation of metabolites

Study cohort and sample collection

This study analyzed serum samples collected from 125 adult participants enrolled at MedStar Georgetown University Hospital and Howard University Hospital under a protocol approved by the Georgetown Institutional Review Board (IRB). Written informed consent and HIPAA authorization were obtained from all individuals prior to participation. Table 1 summarizes the clinical and demographic characteristics of the study groups, which included 50 HCC cases, 50 patients with CIRR, and 25 healthy controls (CTL). Serum analyses were performed using UHPLC coupled with Altis Plus triple quadrupole MS/MS. All HCC patients had a confirmed diagnosis of underlying liver cirrhosis. The diagnosis of HCC was established based on validated imaging criteria and/or histopathological evaluation. Clinical staging of HCC cases was determined according to the tumor-node-metastasis (TNM) classification system. Peripheral venous blood was collected using sterile vacuum tubes without anticoagulants. Samples were centrifuged at room temperature, first at 1000 × g for 10 min, followed by a second centrifugation at 2500 × g for 10 min. The resulting serum was carefully collected, aliquoted, and stored at − 80 °C until further analysis.

Table 1.

Clinical characteristics of study population

		HCC (n = 50)	CIRR (n = 50)	CTL (n = 25)	p-value (HCC vs. CIRR)
Age	Mean (SD)	61 (6.7)	57 (11.6)	53 (8.4)	0.057
Sex	Male	58.0%	54.0%	24.0%	0.502
BMI	Mean (SD)	29.5 (5.1)	30.1 (8.7)	34.2 (12.7)	0.643
Race	EA	44.0%	62.0%	4.0%	0.121
	AA	36.0%	18.0%	96.0%
	Asian	20.0%	20.0%	0.0%
HCV serology	HCV Ab+	60.0%	26.0%	N/A	< 0.001
HBV serology	anti HBC+	30.0%	14.0%	N/A	0.047
	HBs Ag+	16.0%	6.0%	N/A	0.116
Smoking	Current	22.0%	20.0%	36.0%	0.333
	Former	46.0%	34.0%	0.0%
	None	32.0%	46.0%	48.0%
Alcohol	Current	8.0%	18.0%	20.0%	0.304
	Former	52.0%	52.0%	0.0%
	None	38.0%	30.0%	52.0%
MELD	Median (SD)	10.9 (4.9)	17.8 (16.6)	N/A	0.001
AFP	Median (IQR)	16.1 (104)	3.8 (3.3)	N/A	< 0.001
AST	Median (IQR)	78.5 (75.2)	53 (62)	N/A	0.400
ALT	Median (IQR)	63 (48.2)	38 (55.7)	N/A	0.026
HCC stage	Stage I	27	N/A	N/A
	Stage II	11	N/A	N/A
	Stage III	3	N/A	N/A
	Unknown	9	N/A	N/A

Chemicals and reagents

All the internal standards (ISs), including the deuterium ones were purchased from CDN isotopes (Pointe-Claire, QC, Canada), Sigma-Aldrich (St. Louis, MO, USA), and Cambridge Isotope Laboratories (Tewksbury, MA, USA). These include debrisoquine sulfate (DBQ), 4-nitrobenzoic acid (NBA), L-phenyl-d5-alanine-d3, taurodeoxycholic acid, sodium salt (2,2,4,4-d4) (TDCA-d4), glycodeoxycholic-2,2,4,4-d4 acid (GDCA-d4), and arachidonic acid-d8. Standards of targeted metabolites for L-proline, D-ornithine, L-valine, L-leucine, L-isoleucine, DL-phenylalanine, xanthine, glycocholic acid (GCA), GDCA, GUDCA, TCA, taurodeoxycholic acid (TDCA), and taurochenodeoxycholic acid (TCDCA) were purchased from Sigma-Aldrich (St. Louis, MO, USA), Fisher Scientific (Fair Lawn, NJ, USA), and Medchem Express (Monmouth Junction, NJ, USA). Human serum albumin (HAS) and phosphate-buffered saline (PBS) were purchased from Sigma-Aldrich (St. Louis, MO, USA) and Fisher Scientific (Fair Lawn, NJ, USA), respectively. All solvents used for mobile phase preparation and sample extraction, including water, methanol, and acetonitrile, were HPLC grade and sourced from Fisher Scientific (Fair Lawn, NJ, USA) or Honeywell Chemicals (Muskegon, MI, USA).

Preparation of stock and working solutions

The primary stock solutions of all the metabolites and ISs were made at a concentration of 1 mg/mL by dissolving in methanol or water. The working solutions were prepared by diluting the stock solutions using methanol at a concentration of 100 µg/mL. The working solutions for ISs were made by pooling all the ISs in methanol, where the final concentration was 1 µg/mL for L-phenyl-d5-alanine-d3, GDCA-d4, TDCA-d4, and 0.5 µg/mL for DBQ and NBA. All stocks and working solutions were stored at − 20 °C before analysis. On the day of analysis, the calibration standards were prepared by serially diluting the working solutions by methanol at a concentration of 0.78, 1.56, 3.12, 6.25, 12.5, 25, and 50 µg/mL for D-ornithine, L-valine, DL-Phenylalanine, and uric acid, 0.39, 0.78, 1.56, 3.12, 6.25, 12.5, and 25 µg/mL for L-leucine, and L-isoleucine, 0.19, 0.39, 0.78, 1.56, 3.12, 6.25, and 12.5 µg/mL for L-proline, and xanthine. Calibration standards for all six bile acids were prepared by following the same process mentioned above at concentrations of 0.05, 0.1, 0.2, 0.5, 1, 2, and 5 µg/mL. Surrogate matrix was prepared by dissolving 1 g of HAS in 25 mL of PBS solution.

Serum sample preparation

Serum samples were processed using the protein precipitation technique by following a similar method to our untargeted study, with a few modifications (Rashid et al., 2023). Briefly, 20 µL ISs working solution was added to 50 µL of serum. Proteins of the serum samples were precipitated using 250 µL of ice-cold methanol, followed by 1 min of vortex mixing and 10 min of centrifugation at 14,000 rpm at 4 °C. The resultant clear supernatant was then transferred to a new Eppendorf tube and evaporated to dryness using SpeedVac. Reconstitution of the dried residue was performed using 50 µL of 50% methanol in water containing 0.1% formic acid 5 µL was injected for the final analysis. For the preparation of calibration samples, 50 µL standard working solutions for each metabolite were spiked in 50 µL surrogate matrix, and 20 µL ISs working solution was added to it. The extraction of the surrogate matrix was performed by following the same process as that of serum samples mentioned above.

LC-MS/MS conditions

The chromatographic separation of targeted metabolites was carried out with a Vanquish UHPLC system using an ACQUITY UPLC BEH C18 column (2.1 × 100 mm, 1.7 mm, Waters, Milford, MA, USA) by maintaining the column and autosampler temperature at 50 °C and 4 °C, respectively. The mobile phase consisted of 0.1% formic acid in water (v/v; mobile phase A) and in methanol (v/v; mobile phase B) and eluted at a flow rate of 0.3 mL/min following the same gradient as our previous study (Rashid et al., 2023). The chromatographic method was designed to accommodate chemically diverse metabolites in a single run; as expected under reversed-phase conditions, early-eluting polar compounds showed broader peaks while later-eluting analytes were narrower, without affecting reproducibility or peak integration.

Statistical analysis

Univariate statistical analysis of 13 metabolites was performed to identify significantly altered ones in four comparisons: HCC vs. CIRR (50 vs. 50), Stage 1 HCC vs. CIRR (27 vs. 50), HCC vs. CTL (50 vs. 25), and Stage I HCC vs. CIRR (25 vs. 25). For each metabolite, normality within each group was assessed using the Shapiro-Wilk test. When either group met the normality criterion, between-group differences were evaluated using the Welch’s two-sample t-test, otherwise the Mann-Whitney U test was applied. To account for multiple testing across the 13 metabolites, the false discovery rate (FDR) is calculated using the Benjamini-Hochberg procedure.

Receiver operating curves (ROCs) were examined using predicted probabilities from logistic regression models. Adjusted models included metabolite(s) together with age, sex, BMI, MELD score, and HCV status as covariates. The discriminatory performance of a biomarker panel was assessed by the area under the ROC curve (AUC), and incremental value was evaluated by comparing covariate-only and covariate-plus-metabolite models. Metabolite effects were summarized as adjusted odds ratios (ORs) with 95% confidence intervals (CIs), reported per 1-standard-deviation increase in metabolite level to allow comparison across analytes measured on different scales.

Results

Study cohort

A total of 125 participants were included in this study, comprising 50 HCC, 50 CIRR, and 25 CTL. Among the 50 HCC cases, 27 were diagnosed with patients with Stage I HCC. The clinical and demographic characteristics of the study population are summarized in Table 1. No significant differences were observed between HCC and CIRR groups in age, sex, BMI, or race, indicating appropriate comparability between disease groups. However, HCC patients exhibited significantly higher prevalence of HCV positivity (p < 0.001), higher HBV core antibody positivity (p = 0.047), and elevated AFP levels (p < 0.001). In contrast, MELD scores were significantly lower in HCC vs. CIRR (p = 0.001), reflecting differences in liver function severity. ALT levels were also significantly higher in HCC vs. CIRR (p = 0.026).

Analytical performance and method validation

All targeted 13 metabolites were successfully quantified using UHPLC-MS/MS in MRM mode. Calibration curves exhibited high linearity across all analytes, with coefficients of determination (R²) ranging from R² > 0.99, confirming robust quantitative performance. Metabolite identities were rigorously validated using authentic standards based on retention time, precursor/product ion transitions, and MS/MS fragmentation patterns. The use of isotope-labeled and structurally relevant internal standards ensured accurate normalization and minimized analytical variability. The surrogate matrix approach using human serum albumin enabled reliable absolute quantification of endogenous metabolites across a broad dynamic range. A surrogate matrix consisting of HSA in PBS was used for calibration, as the target metabolites are endogenous, and true blank human serum is not available. Table S1 summarizes MRM parameters, internal standards, retention times, calibration ranges, and R² values for all 13 metabolites. Representative extracted MRM chromatograms are shown in Figure S1, while calibration curves for all analytes are provided in Figure S2.

Univariate statistical analysis

In the HCC vs. CIRR comparison, 6 of the 13 targeted metabolites were significantly altered with FDR < 0.1. As shown in Table 2, valine and isoleucine were significantly elevated (FDR < 0.1), whereas phenylalanine, ornithine, GUDCA, and proline were significantly decreased (FDR < 0.1). Fold change (FC) values are also shown in the table. Overall, this analysis identified altered branched-chain amino acids and bile acid species as the main discriminating features between HCC and cirrhosis in this targeted panel. Figure 2 shows the distribution of the selected six targeted metabolites across CTL, CIRR, Stage I HCC, and HCC, demonstrating significant perturbations in amino acid metabolism.

Table 2.

Univariate statistical analysis of 13 metabolites in HCC vs. CIRR and Stage I HCC vs. CIRR (FDR values below 0.1 are shown in bold)

Metabolite	HCC vs. CIRR (50 vs. 50)			Stage I HCC vs. CIRR (27 vs. 50)
	p-value	FDR	FC	p-value	FDR	FC
Phenylalanine	< 0.001	0.007	−1.20	< 0.001	0.003	−1.24
Ornithine	0.037	0.097	−1.13	0.002	0.015	−1.22
Valine	0.004	0.027	1.19	0.021	0.068	1.18
Glycoursodeoxycholic acid (GUDCA)	0.015	0.065	−2.79	0.016	0.068	−9.86
Proline	0.018	0.082	−1.15	0.053	0.137	−1.18
Isoleucine	0.034	0.089	1.13	0.231	0.500	1.07
Leucine	0.109	0.177	1.09	0.426	0.553	1.04
Glycocholic acid (GCA)	0.109	0.177	−1.32	0.352	0.500	−1.22
Taurodeoxycholic acid (TDCA)	0.136	0.179	1.36	0.261	0.500	1.47
Taurochenodeoxycholic acid (TCDCA)	0.137	0.179	−1.36	0.346	0.500	−1.30
Glycodeoxycholic acid (GDCA)	0.211	0.249	−1.42	0.481	0.568	−1.15
Taurocholic acid (TCA)	0.603	0.653	1.11	0.524	0.568	1.23
Xanthine	0.682	0.682	−1.47	0.272	0.500	−1.69

Fig. 2.

Dot plots of six metabolites across CTL, CIRR, Stage I HCC, and HCC with horizontal lines indicating group median. Tables 2 and S2 present the p-values, FDR, and FC for HCC vs. CIRR, Stage I HCC vs. CIRR, HCC vs. CTL, and Stage I HCC vs. CTL comparisons. The HCC group includes Stages I-III and unknowns

In the Stage 1 HCC vs. CIRR comparison, 4 of the 6 metabolites that were found significant in HCC vs. CIRR remained significant with FDR < 0.1 (Table 2). Phenylalanine showed the strongest association and was significantly lower in Stage 1 HCC than in CIRR (FDR = 0.003). Ornithine was also significantly decreased in Stage 1 HCC (FDR = 0.015). Among bile acid species, GUDCA was significantly reduced (FDR = 0.068). In contrast, valine was significantly increased (FDR = 0.068). These findings indicate that metabolic perturbations are detectable even in early-stage HCC and are primarily driven by alterations in amino acid and bile acid pathways.

In the HCC vs. CTL comparison, ten metabolites were significantly altered with FDR < 0.1 (Table S2). The strongest signals were observed for phenylalanine, GCA, TCA, and TCDCA, with additional significant differences in ornithine, GUDCA, TDCA, xanthine, proline, and leucine. With the exception of xanthine, proline, and leucine, the remaining seven metabolites remained significant (FDR < 0.1) in Stage I HCC vs. CTL.

Overall, these findings indicate that bile acid dysregulation and altered amino acid metabolism are detectable in early-stage HCC, while some changes appear stronger in the HCC than in the Stage I HCC. Whereas the HCC vs. CTL comparison captures general HCC-associated metabolic dysregulation, the HCC vs. CIRR comparison is more clinically stringent and highlights markers for tumor-related change beyond liver disease background.

Evaluation of metabolite association with HCC

To assess whether the metabolite signals observed in univariate analyses were independent of underlying clinical differences, we evaluated each metabolite via multivariable logistic regression models adjusted for established clinical covariates MELD score, HCV status, age, sex, and BMI. After adjustment, ornithine remained significantly associated with HCC in both HCC vs. CIRR and Stage I HCC vs. CIRR. In the former cohort, higher ornithine was associated with lower odds of HCC (adjusted OR = 0.44; 95% CI = 0.23–0.83; p = 0.008). In the latter cohort, a similar inverse association was observed (adjusted OR = 0.34; 95% CI = 0.15–0.82; p = 0.016), indicating that this relationship persisted even in early-stage disease. In addition, proline was marginally significantly associated with HCC in the HCC vs. CIRR cohort after covariate adjustment (adjusted OR, 0.50; 95% CI, 0.25–0.99; p = 0.049). No other metabolites retained statistical significance after adjustment in either cohort. Together, these findings suggest that ornithine is the most robust independent metabolite correlate distinguishing HCC from CIRR in this targeted panel, whereas most other observed univariate associations were attenuated after accounting for MELD score, HCV status, age, sex, and BMI.

Evaluation of discriminatory performance

Receiver operating characteristic (ROC) curves were evaluated using logistic regression models with five-fold cross-validation in both the HCC vs. CIRR and Stage I HCC vs. CIRR cohorts. In the unadjusted HCC vs. CIRR analysis, the six-metabolite plus AFP panel showed improved discriminatory performance, with a mean AUC of 0.870 ± 0.083, compared with 0.773 ± 0.106 for AFP alone (DeLong’s test, p = 0.061, FDR = 0.092) (Fig. 3a). A similar pattern was observed in the Stage I HCC vs. CIRR comparison, where the six-metabolite plus AFP panel achieved significantly improved discriminatory power with a mean AUC of 0.877 ± 0.059, compared with 0.804 ± 0.093 for AFP alone (DeLong’s test, p = 0.039, FDR = 0.059) (Fig. 3b). Together, these findings indicate that adding the six-metabolite panel to AFP improves discrimination of both overall HCC and Stage I HCC from cirrhosis.

Fig. 3.

Evaluation in detecting HCC a or Stage I HCC b in patients with CIRR. Receiver operating characteristics curves were obtained by using logistic regression models for discrimination of HCC from CIRR and a five-fold cross-validation for AFP alone, six-metabolite panel, and six-metabolite panel combined with AFP. AUC values were calculated from pooled out-of-fold predictions

After adjustment for age, sex, BMI, MELD score, and HCV status, model discrimination remained unchanged when AFP and metabolite markers were combined with the clinical covariate model. In the HCC vs. CIRR, AFP combined with ornithine and proline had an AUC of 0.828 ± 0.074, while AFP combined with the six-metabolite panel achieved AUC of 0.840 ± 0.077 (Fig. 4a). In the Stage 1 HCC vs. CIRR cohort, the AUC for AFP alone was 0.862 ± 0.092, 0.889 ± 0.068 for AFP plus ornithine and proline, and 0.898 ± 0.054 for AFP plus the six-metabolite panel (Fig. 4b). Overall, the six-metabolite panel combined with AFP provided the best discrimination in both unadjusted and covariate-adjusted analyses, and this pattern was consistent in both the HCC vs. CIRR and the Stage I HCC vs. CIRR cohorts.

Fig. 4.

Evaluation in discriminating of HCC from CIRR a and Stage I HCC from CIRR b after adjustment for age, sex, BMI, MELD score, and HCV status. AUC values were calculated from pooled out-of-fold predictions for AFP alone, ornithine + proline, AFP + ornithine + proline, and AFP + six-metabolite panel via logistic regression models and five-fold cross-validation

At the clinically relevant threshold of 90% specificity, AFP alone in the HCC vs. CIRR cohort had a sensitivity of 56%, whereas the six-metabolite panel and AFP combined had a sensitivity of 50%, despite a higher AUC (0.773 vs. 0.870). In the Stage 1 HCC vs. CIRR cohort, AFP alone had a sensitivity of 59.3% at 92% specificity, while the six-metabolite panel and AFP had a sensitivity of 54.2%, again despite a higher AUC (0.804 vs. 0.877). These findings indicate that the metabolite panel improved overall discrimination, but did not increase sensitivity at a stringent specificity threshold. Also, the positive predictive value (PPV) at the 90%-specificity threshold was 84.8% for AFP alone and 89.3% for the six-metabolite panel and AFP in the HCC vs. CIRR cohort, and 80% vs. 77% in the Stage I HCC vs. CIRR cohort. Assuming a more clinically relevant 20% HCC prevalence, the PPV was 58.3% for AFP alone and 67.6% for the six-metabolite panel and AFP in the HCC vs. CIRR cohort, and 65% vs. 61.2% in the Stage I HCC vs. CIRR cohort. Together, these results suggest that the added metabolite panel may improve overall model discrimination, but do not add value to sensitivity and PPV at the clinically relevant high-specificity threshold.

In covariate-adjusted models, a similar pattern was observed. In the HCC vs. CIRR cohort, sensitivity at 90% specificity was 58% for AFP alone and 60% for AFP plus the six-metabolites panel, although the latter had a slightly higher AUC (0.820 vs. 0.840). In the Stage 1 HCC cohort, AFP had a sensitivity of 55.6% at 90% specificity, compared with 63% for AFP and the six metabolites combined.

Discussion

In this study, we performed targeted quantification of 13 metabolite candidates in serum samples from HCC, CIRR, and CTL subjects using UHPLC-MS/MS in MRM mode. Several metabolites demonstrated significant alterations across groups; however, differences between HCC and CIRR were less pronounced, highlighting the substantial metabolic overlap between these conditions. A major finding of this study is the dysregulation of amino acid metabolism, particularly branched-chain amino acids (BCAAs). BCAAs are known to play essential roles in energy metabolism, nitrogen balance, and hepatocyte proliferation. In the present study, valine and isoleucine were increased in HCC vs. CIRR, while their levels were decreased in CIRR relative to CTL. This pattern is consistent with previous reports demonstrating reduced BCAA levels in CIRR and relative elevation in HCC (Foglia et al., 2023, Zhu et al., 2024, Zhang et al., 2006). The decrease in BCAAs in cirrhosis has been attributed to impaired hepatic synthesis and increased peripheral utilization, particularly for ammonia detoxification (Zhu et al., 2024). In contrast, the elevation observed in HCC may reflect increased metabolic demands of tumor cells and altered anabolic processes associated with hepatocarcinogenesis (Foglia et al., 2023). The observed amino acid levels were generally within reported physiological ranges, with disease-associated differences reflecting relative metabolic alterations rather than absolute deviations (Zhang et al., 2006, Zhu et al., 2024, Morine et al., 2022).

In addition to BCAAs, other amino acids, including proline, ornithine, and phenylalanine, were significantly decreased in HCC vs. CIRR. These findings suggest perturbations in nitrogen metabolism and urea cycle pathways, which are commonly disrupted in liver disease and cancer (Feng et al., 2022, Al-Amodi and Kamel, 2024). Among these, ornithine and proline showed the strongest nominal association with HCC in multivariable analysis (adjusted OR = 0.44 and 0.50, respectively), although statistical significance was not retained after correction for multiple testing, indicating the influence of underlying clinical factors. Bile acid metabolism was also significantly altered. Conjugated bile acids, including GCA, GUDCA, TCA, and TCDCA, were elevated in both HCC and CIRR compared to CTL. This is consistent with previous studies reporting accumulation of bile acids in chronic liver disease due to impaired hepatic clearance and disrupted enterohepatic circulation (Khalil et al., 2021, Thomas et al., 2021, Liu et al., 2019, Colosimo and Tomlinson, 2022, Farhat et al., 2022). Although several bile acids showed directional changes, most differences between HCC and CIRR were not statistically significant and should therefore be interpreted as exploratory trends rather than definitive alterations.

Building on our previous untargeted metabolomics study, which identified GDCA, GUDCA, and TCDCA as potential biomarker candidates,¹⁴ the current targeted analysis expanded the panel to include additional conjugated bile acids (GCA, TCA, and TDCA) based on their reported relevance in HCC. We observed a significant downregulation of GUDCA in HCC vs. CIRR, although other bile acids did not show significant change. These findings are consistent with prior reports showing heterogeneous bile acid profiles in HCC, with both increases and decreases reported depending on disease stage, etiology, and analytical methodology (Thomas et al., 2021, Liu et al., 2019, Colosimo and Tomlinson, 2022). Such variability likely reflects the complex regulation of bile acid synthesis, transport, and signaling pathways during liver disease progression (Colosimo and Tomlinson 2022). Although bile acids showed group-wise differences, they did not meaningfully discriminate HCC from CIRR in this study, and therefore, mechanistic interpretations should be considered exploratory.

Importantly, multivariable logistic regression analysis adjusting for MELD score, HCV status, age, sex, and BMI attenuated the statistical significance of all metabolites after FDR correction. This suggests that metabolite alterations are partially driven by underlying liver function and clinical characteristics rather than being independent markers of HCC. The substantial overlap between cirrhosis and HCC metabolic profiles further complicates biomarker discrimination between these conditions.

Despite these limitations, combining metabolite biomarkers with AFP improved overall diagnostic discrimination. In this analysis, AFP alone achieved an AUC of 0.773 in differentiating HCC from CIRR cohort and 0.804 in the Stage I HCC vs. CIRR comparison. While the six-metabolite panel yielded AUCs of 0.722 and 0.806, respectively, the combined AFP and six-metabolite panel improved the discrimination performance to AUCs of 0.870 in the HCC vs. CIRR and 0.877 in the Stage I HCC vs. CIRR. This suggests that metabolite biomarkers provide complementary information to AFP rather than functioning as superior standalone markers. These findings are consistent with previous studies demonstrating improved diagnostic accuracy using multi-marker approaches (Lu et al., 2022, Morine et al., 2022).

This study represents a targeted validation of metabolite candidates identified in our previous untargeted analysis (Rashid et al., 2023, Ressom et al., 2012). While targeted quantification improves analytical specificity and reproducibility, the assay was not developed as a fully validated bioanalytical method. The findings, therefore, highlight both the potential and the challenges of translating metabolite biomarkers into clinically robust tools for distinguishing HCC from CIRR.

Several limitations should be acknowledged. First, the moderate sample size may have limited statistical power. Second, the metabolic overlap between HCC and CIRR may reduce discriminatory capacity. In addition, this study was designed as an exploratory quantification of selected metabolites using a targeted metabolomics approach and, therefore, did not include full bioanalytical method validation. Although the use of an HSA-in-PBS surrogate matrix enabled preliminary quantitative assessment, key validation parameters, including limit of quantification (LOQ), precision, accuracy, and robustness, were not established. Full validation in accordance with International Council for Harmonisation (ICH) and/or U.S. Food and Drug Administration (FDA) guidelines will be addressed in future studies. Finally, integration of metabolomics with other omics platforms in future work may further enhance biomarker performance and provide deeper insight into the pathogenesis of HCC.

Conclusion

This targeted metabolomics study identified reproducible alterations in amino acid and bile acid metabolism in HCC, including early-stage disease; however, substantial metabolic overlap with cirrhosis limited the independent diagnostic utility of individual metabolites. Among the six metabolites that were significantly altered in HCC vs. CIRR, ornithine and proline showed the most consistent association with HCC after adjusting for clinical covariates. While metabolite panels alone demonstrated moderate performance, their combination with AFP improved overall discrimination in both HCC and Stage I cohorts, although the incremental benefit over AFP alone remained modest. These findings support a complementary rather than standalone role for metabolite biomarkers in HCC detection and highlight the need for larger studies and multi-omics integration to improve diagnostic specificity.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, H.W.R.; methodology, M.M.R; formal analysis, M.M.R. and R.S.V; resources, Z.A.S., A.K, H.W.R.; writing—original draft preparation, M.M.R.; writing—review and editing, M.M.R, R.S.V, M.S.S., Z.A.S., A.K., H.W.R.; visualization, M.M.R., R.S.V.; supervision, H.W.R.; project administration, H.W.R.; funding acquisition, H.W.R. All authors have read and agreed to the published version of the manuscript.

Data availability

Data generated in this work has been deposited in Metabolomics Workbench (Study ID: ST004735) and can be accessed via DOI: http://dx.doi.org/10.21228/M8H84T.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

This study used non-identifiable serum samples collected at MedStar Georgetown University Hospital (IRB 2005 − 206) and Howard University Hospital (IRB 2014-0059), Washington DC. All participating patients provided written informed consent before taking part in the study. Following the participant’s informed consent signature and enrollment, the subject’s blood was drawn, processed, and stored.